FUNDAMENTALS OF DATA MINING IN GENOMICS AND PROTEOMICS FUNDAMENTALS OF DATA MINING IN GENOMICS AND PROTEOMICS Edited by Werner Dubitzky University of Ulster, Coleraine, Northern Ireland Martin Granzow Quantiom Bioinformatics GrmbH & Co KG, Weingarten/Baden, Germany Daniel Berrar University of Ulster, Coleraine, Northern Ireland Springer Library of Congress Control Number: 2006934109 ISBN-13: 978-0-387-47508-0 ISBN-10: 0-387-47508-7 e-ISBN-13: 978-0-387-47509-7 e-ISBN-10: 0-387-47509-5 Printed on acid-free paper © 2007 Springer Science+Business Media, LLC All rights reserved This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in coimection with reviews or scholarly analysis Use in cotmection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden The use in this publication of trade names, trademarks, service marks and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights springer.com Preface As natural phenomena are being probed and mapped in ever-greater detail, scientists in genomics and proteomics are facing an exponentially growing volume of increasingly complex-structured data, information, and knowledge Examples include data from microarray gene expression experiments, bead-based and microfluidic technologies, and advanced high-throughput mass spectrometry A fundamental challenge for life scientists is to explore, analyze, and interpret this information effectively and efficiently To address this challenge, traditional statistical methods are being complemented by methods from data mining, machine learning and artificial intelligence, visualization techniques, and emerging technologies such as Web services and grid computing There exists a broad consensus that sophisticated methods and tools from statistics and data mining are required to address the growing data analysis and interpretation needs in the life sciences However, there is also a great deal of confusion about the arsenal of available techniques and how these should be used to solve concrete analysis problems Partly this confusion is due to a lack of mutual understanding caused by the different concepts, languages, methodologies, and practices prevailing within the different disciplines A typical scenario from pharmaceutical research should illustrate some of the issues A molecular biologist conducts nearly one hundred experiments examining the toxic effect of certain compounds on cultured cells using a microarray gene expression platform The experiments include different compounds and doses and involves nearly 20 000 genes After the experiments are completed, the biologist presents the data to the bioinformatics department and briefly explains what kind of questions the data is supposed to answer Two days later the biologist receives the results which describe the output of a cluster analysis separating the genes into groups of activity and dose While the groups seem to show interesting relationships, they not directly address the questions the biologist has in mind Also, the data sheet accompanying the results shows the original data but in a different order and somehow transformed Discussing this with the bioinformatician again it turns out that what vi Preface the biologist wanted was not clustering {automatic classification or automatic class prediction) but supervised classification or supervised class prediction One main reason for this confusion and lack of mutual understanding is the absence of a conceptual platform that is common to and shared by the two broad disciplines, life science and data analysis Another reason is that data mining in the life sciences is different to that in other typical data mining applications (such as finance, retail, and marketing) because many requirements are fundamentally different Some of the more prominent differences are highlighted below A common theme in many genomic and proteomic investigations is the need for a detailed understanding (descriptive, predictive, explanatory) of genome- and proteome-related entities, processes, systems, and mechanisms A vast body of knowledge describing these entities has been accumulated on a staggering range of life phenomena Most conventional data mining applications not have the requirement of such a deep understanding and there is nothing that compares to the global knowledge base in the hfe sciences A great deal of the data generated in genomics and proteomics is generated in order to analyze and interpret them in the context of the questions and hypotheses to be answered and tested In many classical data mining scenarios, the data to be analyzed axe generated as a "by-product" of an underlying business process (e.g., customer relationship management, financial transactions, process control, Web access log, etc.) Hence, in the conventional scenario there is no notion of question or hypothesis at the point of data generation Depending on what phenomenon is being studied and the methodology and technology used to generate data, genomic and proteomic data structures and volumes vary considerably They include temporally and spatially resolved data (e.g., from various imaging instruments), data from spectral analysis, encodings for the sequential and spatial representation of biological macromolecules and smaller chemical and biochemical compounds, graph structures, and natural language text, etc In comparison, data structures encountered in typical data mining applications are simple Because of ethical constraints and the costs and time involved to run experiments, most studies in genomics and proteomics create a modest number of observation points ranging from several dozen to several hundreds The number of observation points in classical data mining applications ranges from thousands to millions On the other hand, modern high-throughput experiments measure several thousand variables per observation, much more than encountered in conventional data mining scenarios By definition, research and development in genomics and proteomics is subject to constant change - new questions are being asked, new phenomena are being probed, and new instruments are being developed This leads to frequently changing data processing pipelines and workflows Business processes in classical data mining areas are much more stable Because solutions will be in use for a long time, the development of complex, comprehensive, and Preface vii expensive data mining applications (such as data warehouses) is readily justified Genomics and proteomics are intrinsically "global" - in the sense that hundreds if not thousands of databases, knowledge bases, computer programs, and document libraries are available via the Internet and are used by researchers and developers throughout the world as part of their day-to-day work The information accessible through these sources form an intrinsic part of the data analysis and interpretation process No comparable infrastructure exists in conventional data mining scenarios This volume presents state of the art analytical methods to address key analysis tasks that data from genomics and proteomics involve Most importantly, the book will put particular emphasis on the common caveats and pitfalls of the methods by addressing the following questions: What are the requirements for a particular method? How are the methods deployed and used? When should a method not be used? What can go wrong? How can the results be interpreted? The main objectives of the book include: • • • • • • • • • To be acceptable and accessible to researchers and developers both in life science and computer science disciplines - it is therefore necessary to express the methodology in a language that practitioners in both disciplines understand; To incorporate fundamental concepts from both conventional statistics as well as the more exploratory, algorithmic and computational methods provided by data mining; To take into account the fact that data analysis in genomics and proteomics is carried out against the backdrop of a huge body of existing formal knowledge about life phenomena and biological systems; To consider recent developments in genomics and proteomics such as the need to view biological entities and processes as systems rather than collections of isolated parts; To address the current trend in genomics and proteomics towards increasing computerization, for example, computer-based modeling and simular tion of biological systems and the data analysis issues arising from largescale simulations; To demonstrate where and how the respective methods have been successfully employed and to provide guidelines on how to deploy and use them; To discuss the advantages and disadvantages of the presented methods, thus allowing the user to make an informed decision in identifying and choosing the appropriate method and tool; To demonstrate potential caveats and pitfalls of the methods so as to prevent any inappropriate use; To provide a section describing the formal aspects of the discussed methodologies and methods; viii • • Preface To provide an exhaustive list of references the reader can follow up to obtain detailed information on the approaches presented in the book; To provide a list of freely and commercially available software tools It is hoped that this volume will (i) foster the understanding and use of powerful statistical and data mining methods and tools in life science as well as computer science and (ii) promote the standardization of data analysis and interpretation in genomics and proteomics The approach taken in this book is conceptual and practical in nature This means that the presented dataranalytical methodologies and methods are described in a largely non-mathematical way, emphasizing an informationprocessing perspective (input, output, parameters, processing, interpretation) and conceptual descriptions in terms of mechanisms, components, and properties In doing so, the reader is not required to possess detailed knowledge of advanced theory and mathematics Importantly, the merits and limitations of the presented methodologies and methods are discussed in the context of "real-world" data from genomics and proteomics Alternative techniques are mentioned where appropriate Detailed guidelines are provided to help practitioners avoid common caveats and pitfalls, e.g., with respect to specific parameter settings, sampling strategies for classification tasks, and interpretation of results For completeness reasons, a short section outlining mathematical details accompanies a chapter if appropriate Each chapter provides a rich reference list to more exhaustive technical and mathematical literature about the respective methods Our goal in developing this book is to address complex issues arising from data analysis and interpretation tasks in genomics and proteomics by providing what is simultaneously a design blueprint, user guide, and research agenda for current and future developments in the field As design blueprint, the book is intended for the practicing professional (researcher, developer) tasked with the analysis and interpretation of data generated by high-throughput technologies in genomics and proteomics, e.g., in pharmaceutical and biotech companies, and academic institutes As a user guide, the book seeks to address the requirements of scientists and researchers to gain a basic understanding of existing concepts and methods for analyzing and interpreting high-throughput genomics and proteomics data To assist such users, the key concepts and assumptions of the various techniques, their conceptual and computational merits and limitations are explained, and guidelines for choosing the methods and tools most appropriate to the analytical tasks are given Instead of presenting a complete and intricate mathematical treatment of the presented analysis methodologies, our aim is to provide the users with a clear understanding and practical know-how of the relevant concepts and methods so that they are able to make informed and effective choices for data preparation, parameter setting, output postprocessing, and result interpretation and validation Preface ix As a research agenda, this volume is intended for students, teachers, researchers, and research managers who want to understand the state of the art of the presented methods and the areas in which gaps in our knowledge demand further research and development To this end, our aim is to maintain the readability and accessibility throughout the chapters, rather than compiling a mere reference manual Therefore, considerable effort is made to ensure that the presented material is supplemented by rich literature cross-references to more foundational work In a quarter-length course, one lecture can be devoted to two chapters, and a project may be assigned based on one of the topics or techniques discussed in a chapter In a semester-length course, some topics can be covered in greater depth, covering - perhaps with the aid of an in-depth statistics/data mining text - more of the formal background of the discussed methodology Throughout the book concrete suggestions for further reading are provided Clearly, we cannot expect to justice to all three goals in a single book However, we beheve that this book has the potential to go a long way in bridging a considerable gap that currently exists between scientists in the field of genomics and proteomics on one the hand and computer scientists on the other hand Thus, we hope, this volume will contribute to increased communication and collaboration across the disciplines and will help facilitate a consistent approach to analysis and interpretation problems in genomics and proteomics in the future This volume comprises 12 chapters, which follow a similar structure in terms of the main sections The centerpiece of each chapter represents a case study that demonstrates the use - and misuse - of the presented method or approach The first chapter provides a general introduction to the field of data mining in genomics and proteomics The remaining chapters are intended to shed more light on specific methods or approaches The second chapter focuses on study design principles and discusses replication, blocking, and randomization While these principles are presented in the context of microarray experiments, they are applicable to many types of experiments Chapter addresses data pre-processing in cDNA and oligonucleotide microarrays The methods discussed include background intensity correction, data normalization and transformation, how to make gene expression levels comparable across different arrays, and others Chapter is also concerned with pre-processing However, the focus is placed on high-throughput mass spectrometry data Key topics include baseline correction, intensity normalization, signal denoising (e.g., via wavelets), peak extraction, and spectra alignment Data visualization plays an important role in exploratory data analysis Generally, it is a good idea to look at the distribution of the data prior to analysis Chapter revolves around visualization techniques for highdimensional data sets, and puts emphasis on multi-dimensional scaling This technique is illustrated on mass spectrometry data X Preface Chapter presents the state of the art of clustering techniques for discovering groups in high-dimensional data The methods covered include hierarchical and fc-means clustering, self-organizing maps, self-organizing tree algorithms, model-based clustering, and cluster validation strategies, such as functional interpretation of clustering results in the context of microarray data Chapter addresses the important topics of feature selection, feature weighting, and dimension reduction for high-dimensional data sets in genomics and proteomics This chapter also includes statistical tests (parametric or nonparametric) for assessing the significance of selected features, for example, based on random permutation testing Since data sets in genomics and proteomics are usually relatively small with respect to the number of samples, predictive models are frequently tested based on resampled data subsets Chapter reviews some common data resampling strategies, including n-fold cross-validation, leave-one-out crossvalidation, and repeated hold-out method Chapter discusses support vector machines for classification tasks, and illustrates their use in the context of mass spectrometry data Chapter 10 presents graphs and networks in genomics and proteomics, such as biological networks, pathways, topologies, interaction patterns, gene-gene interactome, and others Chapter 11 concentrates on time series analysis in genomics A methodology for identifying important predictors of time-varying outcomes is presented The methodology is illustrated in a study aimed at finding mutations of the human immunodeficiency virus that are important predictors of how well a patient responds to a drug regimen containing two different antiretroviral drugs Automated extraction of information from biological literature promises to play an increasingly important role in text-based knowledge discovery processes This is particularly important for high-throughput approaches such as microarrays and high-throughput proteomics Chapter 12 addresses knowledge extraction via text mining and natural language processing Finally, we would like to acknowledge the excellent contributions of the authors and Alice McQuillan for her help in proofreading Coleraine, Northern Ireland, and Weingajten, Germany Werner Dubitzky Martin Granzow Daniel Berrar 268 Robert Hoffmann Table 12.2 Text mining tools and resources in biology - information extraction Source Description and U R L BioIE Rule-based system that extracts informative sentences from PubMed query results http://umber.sbs.man.ac.uk/dbbrowser/bioie/ Queries the biomedical literature for specific entity relationships, h t t p : / / t e x t m i n e c u - g e n o m e o r g / gridsphere/gridsphere Protein annotation and tagging, http://pir.georgetown.edu/lprolink Gene-to-gene co-citation network that can be used for microarray analysis, http://www.pubgene.org Annotate proteins from scientific references, h t t p : / / www.bork.embl-heidelberg.de/kat/index.html JournalMine iProLINK PubGene KAT Data integration TxtGate STRING Entity recognition ABNER GAPSCORE NLProt Summarization and analysis of groups of genes based on text, h t t p : / / t o m c a t e s a t k u l e u v e n b e : 8080/txtgate/home.j sp Integration of protein interaction extracted from the literature with information from complementary methods, h t t p : / / s t r i n g e m b l d e Entity detection h t t p : //www OS wise, edu/''"bsettles/abner Protein gene name tagger http://bionlp.Stanford.edu/gapscore Protein/gene name tagger http://rostlab.org/services/nlprot/ Relationship extraction tool, http://www.chilibot.net/ Protein interactions Data mining tool that helps researchers locate bioChilibot molecular interaction information in the scientific litPreBIND erature, h t t p : / / p r e b i n d b i n d c a Knowledge discovery Arrowsmith A tool for identifying links between two sets of PubMed articles, h t t p : //arrowsmith psych u i c edu BITOLA Aims to facilitate the discovery of potentially new relations between biomedical concepts http://www.mf.uni-lj.si/bitola HCAD Provides comprehensive information on human chromosomal aberrations, including genes and disease rela,tionships h t t p : //www i h o p - n e t org/UniPub/HCAD/ G2D Finds literature links between OMIM entries and genes from a specific chromosomal location h t t p : / / w w w o g i c c a / p r o j ects/g2d-2 12 Text Mining in Genomics and Proteomics 269 T a b l e 12.3 Text mining tools and resources in biology - annotated text corpora Source Description and U R L BioCreative corpus Corpus of protein annotation relevant text http://www.pdg.cnb.uam.es/BioLINK/ h t t p : / / f e t c h p r o t s i c s se f t p : / / f t p n c b i nlm n i h gov/pub/tanabe Annotated corpus related to human blood transcrij)tion factors http://www-tsuj i i i s s u - t o k y o a c j p / G E N I A h t t p : / / b i o i e I d c upenn edu http://www.sics.se/humle/proj ects/prothalt FetchProt GENETAG GENIA PennBioIE Yapex Assessments BioCreative Challenge Text mining of protein names and annotations http://www.pdg.cnb.uam.es/BioLINK/ BioCreative.eval.html KDD challenge Information extraction of Drosophila gene expression information h t t p : / / w w w b i o s t a t w i s e edu/~craven/kddcup/ tasks.html TREC Genomics track IR, document classification and question answering http://ir.ohsu.edu/genomics/ P a r t - o f - s p e e c h t a g g e r s Marking up the words in a text with their corresponding parts of speech (e.g., verbs, nouns) Brill h t t p : //www c s j h u e d u / ~ b r i l l TNT h t t p : //www c o l i u n i - s a a r l a n d d e / ~ t h o r s t e n / t n t TreeTagger h t t p : //www ims u n i - s t u t t g e i r t de/~schmid N a t Lzinguage P a r s e r s Derive the grammatical structure of sentences, e.g., which groups of words are units (phrases) and which words are the subject or object of a verb CASS http://www.vinartus.net/spa Collins Parser h t t p : / / p e o p l e c s a i l mit edu/mcollins Stanford Parser http://nlp.stanford.edu/software of genome-wide data: Methods t o asses the coherence of gene groups (Raychaudhuri et al., 2003), t o integrate experimental d a t a with literature networks (Jenssen et al., 2001; HofTmann and Valencia, 2005; von Mering et al., 2005) and ways t o make a simultaneous analysis of literature and experimental d a t a possible (Hoffmann and Valencia, 2004) Many of these methods have led t o t h e development of tools and Web sites ready t o use Other promising methods are still in an experimental phase, b u t will soon reach productionstate Hence, developers of novel analysis software and workflows are able t o choose from a variety of stable text mining solutions T h e recent development in science towards open and freely accessible full text-resources will further catalyze this progress However, I have also discussed some of t h e important difficulties and caveats t h a t text mining methods are still facing in biology: T h e vast number 270 Robert Hoffmann of ambiguous acronyms and symbols and the complexity of scientific language Addressing these problems at a pragmatic level is important, but it cannot be an aim on its own Not in the light of gigabytes of data to be expected from large-scale experiments in the near future Text mining in biology has to focus on biology-driven problems to maintain the momentum gained over the past decade Thus, some problems that are due to the complexity of natural language might be neglected, but these deficits will be more than compensated by the integration with independently derived sources of information (e.g., large scale experimental data and in silico predictions), as pioneered in recent years by a number of groups (Jenssen et al., 2001; Hoffmann and Valencia, 2004; von Mering et al., 2005) Following this direction, text mining in biology will live up to its full potential and will become an integral element of all future approaches to analyze and interpret novel data 12.9 M a t h e m a t i c a l Details To assess the content of a given document cluster, one can compare the frequencies of scientific terms within the cluster to their frequencies in a reference cluster (e.g., all documents) The probability (PT) of finding a term (T) the observed number of times (fc) in a document cluster (C) is then calculated from the Poisson distribution, given the known reference frequency (p) and the total number of terms in the cluster (n) This approximation is valid when the total number of terms in the reference cluster is much greater than n and p is small P^(k\n,p)=e-^^^ (12.2) where n e N, the number of terms assigned to a document cluster (C), k = 1,2, n, the number of occurrences of term (T) within the cluster (C), p is the relative frequency of term (T) in the reference cluster In practice the log probability can be calculated to avoid floating point errors and n\ can be estimated using Stirling's approximation for large n: lnPT{k\n,p) = -np + kln{np) + k-kln{k) (12.3) References Al-Shahrour, F., Diaz-Uriarte, R., and Dopazo, J (2004) FatiGO: A web tool for finding significant associations of gene ontology terms with groups of genes Bioinformatics, 20(4):578-580 12 Text Mining in Genomics and Proteomics 271 Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M., Davis, A.P., Dolinski, K., Dwight, S.S., Eppig, J.T., Haxris, M.A., Hill, D.P., Issel-Tarver, L., Kasarskis, A., Lewis, S., Matese, J.C., Richardson, J.E., Ringwald, M., Rubin, G.M., and Sherlock, G (2000) Gene ontology: Tool for the unification of biology, the gene ontology consortium Nat Genet., 25(l):25-29 Blaschke, C , Leon, E A., Krallinger, M., and Valencia, A (2005) Evaluation of BioCreAtlvE assessment of task BMC Bioinformatics, Suppl Blaschke, C., Oliveros, J C., and Valencia, A (2001) Mining functional information associated with expression arrays Functional and Integrative Genomics, 1(4):256 Blaschke, C and Valencia, A (2001) The potential use of SUISEKI as a protein interaction discovery tool Genome informatics series: Proc Workshop on Genome Informatics, 12:123 Briscoe, T and Carroll, J (2002) Robust accurate statistical annotation of general text Proc 3rd Intl Conf Language Resources and Evaluation, pages 1499-1504 Chaussabel, D and Sher, A (2002) Mining microarray expression data by literature profihng Genome Biol, 3(10):RESEARCH0055 Collier, N., Nobata, C , and Tsujii, J (2000) Extracting the names of genes and gene products with a hidden markov model Proc COLING 2000, pages 201-207 Cooper, J.W and Kershenbaum, A (2005) Discovery of protein-protein interactions using a combination of linguistic, statistical and graphical information BMC Bioinformatics, 6(1): 143 DeRisi, J.L., Iyer, V.R., and Brown, R O (1997) Exploring the metabolic and genetic control of gene expression on a genomic scale Science, 278(5338) :680-686 Donaldson, L, Martin, J., de Bruijn, B., Wolting, C , Lay, V., Tuekam, B., Zhang, S., Baskin, B., Bader, G.D., Michalickova, K., Pawson, T., and Hogue, C.W (2003) Prebind and textomy-mining the biomedical literar ture for protein-protein interactions using a support vector machine BMC Bioinformatics, 4:11 Pranzen, K., Eriksson, G., Olsson, F., Asker, L., Liden, P., and Coster, J (2002) Protein names and how to find them Int J Med Inf, 67(l-3):4961 Priedl, J.E.F (2002) Mastering regular expressions O'Reilly, Sebastopol, 2nd edition Friedman, C , Kra, P., Yu, H., Krauthammer, M., and Rzhetsky, A (2001) GENIES: A natural-language processing system for the extraction of molecular pathways from journal articles Bioinformatics, 17 Suppl l:S74-82 Fukuda, K., Tamura, A., Tsunoda, T., and Takagi, T (1998) Toward information extraction: Identifying protein names from biological papers Pac Symp Biocomput, pages 707-718 272 Robert Hoffmann Glenisson, P., Coessens, B., Van Vooren, S., Mathys, J., Moreau, Y., and De Moor, B (2004) Txtgate: profiling gene groups with text-based information Genome Biol., 5(6):R43 Hanisch, D., Fluck, J., Mevissen, H T., and Zimmer, R (2003) Playing biology's name game: Identifying protein names in scientific text Pac Symp Biocomp., pages 403-14 Hausser, R.R (2001) Foundations of Computational Linguistics: HumanComputer Communication in Natural Language Springer, Berlin/New York, 2nd edition Heim, S and Mitelman, F (1995) Cancer Cytogenetics Wiley-Liss, New York, 2nd edition Hirschman, L., Morgan, A.A., and Yeh, A.S (2002) Rutabaga by any other name: Extracting biological names J Biomed Inform, 35(4):247-59 Hirschman, L., Yeh, A., Blaschke, C , and Valencia, A (2005) Overview of biocreative: Critical assessment of information extraction for biology BMC Bioinformatics, Suppl Hoffmann, R., Dopazo, J., Cigudosa, J C , and Valencia, A (2005) HCAD, closing the gap between breakpoints and genes Nucleic Acids Res., 33(Database issue):D511-D513 Hoffmann, R and Valencia, A (2003) Life cycles of successful genes Trends Genet, 19(2):79-81 Hoffmann, R and Valencia, A (2004) A gene network for navigating the literature Nat Genet., 36(7):664 Hoffmann, R and Valencia, A (2005) Implementing the iHOP concept for navigation of biomedical literature Bioinformatics, 21 Suppl 2:ii252-ii258 Jensen, L.J., Saric, J., and Bork, P (2006) Literature mining for the biologist: from information retrieval to biological discovery Nat Rev Genet., 7(2):119-129 Jenssen, T.K., Laegreid, A., Komorowski, J., and Hovig, E (2001) A literature network of human genes for high-throughput analysis of gene expression Nat Genet, 28(l):21-28 Kim, J.D., Ohta, T., Tateisi, Y., and Tsujii, J (2003) GENIA corpus-A semantically annotated corpus for bio-textmining Bioinformatics, 19 Suppl 1:1180-1182 Kim, W., Aronson, A.R., and Wilbur, W.J (2001) Automatic MeSH term assignment and quality assessment Proc AMIA Symp., pages 319-23 Krauthammer, M., Rzhetsky, A., Morozov, P., and Friedman, C (2000) Using BLAST for identifying gene and protein names in journal articles Gene, 259(1-2) :245-252 Kuffner, R., Fundel, K., and Zimmer, R (2005) Expert knowledge without the expert: Integrated analysis of gene expression and literature to derive active functional contexts Bioinformatics, 21 Suppl 2:ii259-ii267 Lander, E.S., Linton, L.M., and Birren, B., et al (2001) Initial sequencing and analysis of the human genome Nature, 409(6822) :860-921 12 Text Mining in Genomics and Proteomics 273 Liu, F., Jenssen, T.K., Nygaard, V., Sack, J., and Hovig, E (2004) FigSearch: A figure legend indexing and classification system Bioinformatics, 20(16):2880-2882 Marcotte, E.M., Xenarios, I., and Eisenberg, D (2001) Mining literature for protein-protein interactions Bioinformatics, 17(4):359-363 Masys, D.R., Welsh, J.B., Lynn Fink, J., Gribskov, M., Klacansky, L, and Corbeil, J (2001) Use of keyword hierarchies to interpret gene expression patterns Bioinformatics, 17(4):319-326 Mi, H., Vandergriff, J., Campbell, M., Narechania, A., Majoros, W., Lewis, S., Thomas, P D., and Ashburner, M (2003) Assessment of genome-wide protein function classification for Drosophila melanogaster Genome Res., 13(9):2118-2128 Mika, S and Rost, B (2004) Protein names precisely peeled off free text Bioinformatics, 20 Suppl 1:1241-1247 Mitelman, F., Mertens, F., and Johansson, B (1997) A breakpoint map of recurrent chromosomal rearrangements in human neoplasia Nat Genet., 15 Spec No.:417-474 Morgan, A., Hirschman, L., Yeh, A., and Colosimo, M (2003) Gene name extraction using FlyBase resources ACL-03 Workshop on Natural Language Processing in Biomedicine, pages 1-8 NLM (2006) Yearly citation count totals US National Library of Medicine http://www.nlm.nih.gov Ono, T., Hishigaki, H., Tanigami, A., and Takagi, T (2001) Automated extraction of information on protein-protein interactions from the biological literature Bioinformatics, 17(2):155-161 Park, J.C., Kim, H.S., and Kim, J.J (2001) Bidirectional incremental parsing for automatic pathway identification with combinatory categorial grammar Pac Symp Biocomp., pages 396-407 Phizicky, E., Bastiaens, P.L, Zhu, H., Snyder, M., and Fields, S (2003) Protein analysis on a proteomic scale Nature, 422(6928):208-215 Proux, D., Rechenmann, F., Julliard, L., Pillet, V.V., and Jacq, B (1998) Detecting gene symbols and names in biological texts: A first step toward pertinent information extraction Genome Inform Ser Workshop Genome Inform., 9:72-80 Rabbitts, T.H (1994) Chromosomal translocations in human cancer Nature, 372(6502):143-149 Raychaudhuri, S., Chang, J.T., Imam, F., and Altman, R.B (2003) The computational analysis of scientific literature to define and recognize gene expression clusters Nucleic Acids Res., 31(15):4553-4560 Raychaudhuri, S., Schutze, H., and Altman, R.B (2002) Using text analysis to identify functionally coherent gene groups Genome Res., 12(10):15821590 Schuemie, M.J., Weeber, M., Schijvenaars, B.J., van Mulligen, E.M., van der Eijk, C.C, Jelier, R., Mons, B., and Kors, J.A (2004) Distribution of infor- 274 Robert Hoffmann mation in biomedical abstracts and full-text publications Bioinformatics, 20(16):2597-2604 Shah, P.K., Perez-Iratxeta, C , Bork, P., and Andrade, M.A (2003) Informar tion extraction from full text scientific articles: Where are the keywords? BMC Bioinformatics, 4:20 Shatkay, H., Edwards, S., Wilbur, W J., and Boguski, M (2000) Genes, themes and microarrays: using information retrieval for large-scale gene analysis Proc Intl Conf Intell Syst Mol Biol, 8:317-328 Sherlock, G (2000) Analysis of large-scale gene expression data Curr Opin Immunol, 12(2):201-205 Stuart, J.M., Segal, E., Koller, D., and Kim, S.K (2003) A gene-coexpression network for global discovery of conserved genetic modules Science, 302 (5643) :249-255 Tamames, J and Valencia, A (2006) The success (or not) of HUGO nomenclature Genome Biology, in press Tanabe, L., Scherf, U., Smith, L.H., Lee, J.K., Hunter, L., and Weinstein, J.N (1999) MedMiner: An internet text-mining tool for biomedical information, with application to gene expression profiling Biotechniques, 27(6):12101214, 1216-1217 Tsuruoka, Y and Tsujii, J (2003) Boosting precision and recall of dictionarybased protein name recognition ACL-03 Workshop on Natural Language Processing in Biomedicine, pages 1-8 Vogelstein, B and Kinzler, K.W (2002) The Genetic Basis of Human Cancer McGraw-Hill Medical Pub Division, New York, 2nd edition von Mering, C., Jensen, L.J., Snel, B., Hooper, S.D., Krupp, M., Foglierini, M., Jouffre, N., Huynen, M.A., and Bork, P (2005) STRING: Known and predicted protein-protein associations, integrated and transferred across organisms Nucleic Acids Res., 33(Database issue):D433-437 White, J.A., McAlpine, P.J., Antonarakis, S., Cann, H., Eppig, J.T., Prazer, K., Prezal, J., Lancet, D., Nahmias, J., Pearson, P., Peters, J., Scott, A., Scott, H., Spurr, N., Talbot, C., Jr., and Povey, S (1997) Guidelines for human gene nomenclature (1997) HUGO Nomenclature Committee Genomics, 45(2):468-471 Witten, LH., Moffat, Alistair, and Bell, Timothy C (1999) Managing gigabytes: Compressing and indexing documents and images Morgan Kaufmann Series in Multimedia Information and Systems Morgan Kaufmann Publishers, San Francisco, Calif., 2nd edition Yu, H., Hatzivassiloglou, V., Rzhetsky, A., and Wilbur, W J (2002) Automatically identifying gene/protein terms in medline abstracts J Biomed Inform., 35(5-6):322-330 Zeeberg, B.R., Qin, H., and Narasimhan, S et al (2005) High-throughput GoMiner, an 'industrial-strength' integrative gene ontology tool for interpretation of multiple-microarray experiments, with application to studies of Common Variable Immune Deficiency (CVID) BMC Bioinformatics, 6:168 Index X , see Chi square 2D-DIGE, see two-dimensional difference in-gel electrophoresis 2D-PAGE, see two-dimensional polyaxirylamide gel electrophoresis accuracy, 24 balanced, 24, 194 adjacency matrix, 206-207 affycomp, 64 Affymetrix, 5, 52-53, 64 Agilent, 53-55 amplicon, ANOVA, see one-way analysis of variance arabadopsis, 47 ARD, see automatic relevance determination area under the curve, 152, 169 ArrayAssist, 62 assortative mixing, 210, 220 assortativenness, 211 AUG, see area under the curve automatic relevance determination, 156 BACG, see accuracy, balanced background correction, 55 backward elimination, 17, 155 bagging, see bootstrap aggregation baseline subtraction, 82 beam search, 155 betweenness centrality, 208 distribution, 210 bias, 22 experimenter, 14 selection, 27, 177, 196 bias-variance traxie-off, 23 biclustering, 124, 131, 135 Bioconductor, 62, 74, 98, 104, 112, 120 blocking, 43, 47 complete block design, 43 blotting Northern, Southern, Western, Bonferroni, 20, 134, 154, 232 bootstrap, 28, 179 632 bootstrap, 28, 179 632+ bootstrap, 180, 184 aggegration, 238 BRB-ArrayTools, 184 GAAT, 138 caBig, 98 calibrant, 84, 119 calibration, 81, 82 capaxiity control, 23 GART, see classification and regression trees centroid, 193 Ghi square, 152, 169 Giphergen, 84, 93, 161 GlaNC, 197 class comparison, 41, 124 class discovery, 40, 124 class prediction, 124 276 Index classification, 10 classification and regression trees, 155, 166 classifier lineax majdmal margin, 189 non-linear, 188 CLENCH, 134 cluster analysis, 40 clustering, 10, 124, 156, 158 average, 209 biclustering, 131 coefficient, 207 distribution, 210 fuzzy, 137 hierarchical, 46, 126, 128 average linkage, 129 complete linkage, 129 single linkage, 129 k-means, 126, 129, 158 model-based, 130 probabilistic, 158 self-organizing maps, 130 self-organizing tree algorithm, 130 soft, 158 confidence interval, 24 confounding, 40, 47 connectedness, 131 connectivity, see node, degree correlation filtering, 154 jackknifed correlation coefficient, 127 Pearson, 117, 126 profile, 210 Spearman rank, 127 covariate, cross-hybridization, 53 cross-validation, 27, 191 10-fold, 184 5x2CV, 30 external, 27, 181 internal, 27, 181 k-fold, 28, 178 leave-k-out, 28 leave-one-out, 28, 177, 184 curse of dimensionality, Cy3, 52 Cy5, 52 D / S / A algorithm, 230 data mining, 2, data re-scaling, 15 data transformation, 14 Daubechies wavelet, 87 DAVID, 134 dChip MBEI, 63 DDWT, see decimated discrete wavelet trajisform decimated discrete wavelet transform, 86 deconvolution, 92 dendrogram, 10, 104 detector, differential display, discrete wavelet transform, 86 discriminant analysis, 41 distance metric, 135, 157 City Block, see distance metric, Manhattan correlation, 157 Cosine, 157 Euclidean, 105, 126, 157 standardized, 157 Hamming, 157 Jax;card, 157 Mahalanobis, 137, 157 Manhattan, 112, 157 Minkowski, 157 Dunn-like indices, 131 dye-swap, 45, 47 EAM, see energy absorbing matrix eGOn, 134 eigengene, 16 electrophoresis, electrospray ionization, embedded methods, 150, 155, 160 energy absorbing matrix, 79 entity recognition, 254 ER graph, see graph, Erdos-Renyi error of prediction, 25 rate, 24 family-wise, 20 comparison-wise, 20 observed, 24-25, 181, 183 true, 24-25, 181 resubstitution estimate, 174 selection, 177 Index split-sample estimate, 175 Type I, 18, 258 Type II, 19, 258 ESI, see electrospray ionization EST, see expressed sequence tag ETA, see experimental treatment assignment experimental treatment assignment assumption, 235 expressed sequence tag, F-measure, 258 false discovery rate, 20, 97, 134, 154, 233 false positive rate, 20 FatiGO, 134 FatiGOplus, 138 FDR, see false discovery rate feature, 7, 14-9 construction, 151 selection, 149-169, 190 recursive feature elimination, 190 filter, 150, 160, 169, 190 fingerprint, Fisher score, 151, 152, 169 Fisher-like score, 18 FLD, see linear discriminant, Fisher forward selection, 16, 155 FPR, see false positive rate FWER, see error rate, fajnily-wise Gaussian mixture, 158 GCRMA, 63 Gene Expression Omnibus, 53, 65, 69 Gene Ontology, 261 GeneChip, 5, 52 GenePix, 55, 60, 69 GeneSifter, 62 GeneSpring, 62 genetic algorithm, 150, 155 genomics, functional, GEO, see Gene Expression Omnibus GO, see Gene Ontology GoMiner, 134 goodness-of-fit, 30, 111 GOStat, 134 Gosurfer, 134 GOTM, 134 277 GOToolBox, 134 graph, 203, 206 directed, 205 Erdos-Renyi, 204, 207 k-scaffold, 211 random, 209 random modular graph, 204 scale-free, 204, 209 undirected, 206 weighted, 206 Graphviz, 219 heatmap, 104 Hidden Markov model, 125 high-throughput, hill climbing, 150, 155 hyperplane, 200 maximal margin, 188 hyperplanes, 188 ICA, see independent component analysis IE, see information extraction iHOP, 252, 254, 260 Imagene, 55 Incogen, 99 independent component analysis, 16, 159 inference, 43 information extraction, 254, 268 retrieval, 253, 267 interquartile range, 59 inverse-probability-of-treatmentweighted transformation, 231 ion source, IPTW, see inverse-probability-oftreatment-weighted transformer tion IQR, see interquartile range IR, see information retrieval iterative signature algorithm, 131 J5-score, 152, 169 ja,ckknife, 178 k-nearest neighbor, 26 k-NN, see k-nearest neighbor Karhunen-Loeve transform, see principal component analysis 278 Index kernel function, 189, 200 Ll-metric, see distance metric, Manhattan L2-metric, see distance metric, Euclidean Lagrangian, 199 latent vectors, 16 learning supervised, 40-41 unsupervised, 9, 40 learning by rote, 24 LIBSVM, 197 lift, 24 linear discriminant Fisher, 159 loess, 60, 245 print-tip, 61 LOOCV, see cross-validation, leave-one-out28 m/z, see mass-to-charge ratio MA-plot, 60-61, 76 MAC, see maximum allowed absolute correlation MALDI, see matrix-assisted laser desorption/ionization margin, 198 Markov blanket filtering, 17 MAS 5.0, 64 mass analyzer, mass spectrometry, 5, 79-99, 180 mass-to-chaxge ratio, 5, 80 MATLAB, 98 matrix-assisted laser desorption and ionization, 79 matrix-assisted laser desorption/ionization, maximum allowed absolute correlation, 164 MDS, see multidimensional scaling Medical Subject Headings, 252 Medline, 255 MeSH, see Medical Subject Headings microarray, 5, 39, 51-76 cDNA, 53, 55 single-channel, 52 spotted cDNA array, two-channel, 52, 59 microarray sample pool, 64 mismatch, 54 missing value handling, 13 MM, see mismatch model, assessment, 173, 191 construction, 182 selection, 173, 181, 182 modifications posttranslational, 1, modularity, 203 Monte Carlo cross-validation, see sampling, repeated random subsampling Monte Carlo permutation, 21 MS, see mass spectrometry MSP, see microarray sample pool MUDWT, 96, 97 multi-array probe-level model, 57 multidimensional scaling, 12, 105, 159 multiple hypotheses testing, 19 mutual information, 152, 169 mzXML, 99 natural language processing, 255 negative predictive value, 24 neighborhood divergence, 259 network, 203 cellular, 204 hierarchical, 210 ii, 206 motif, 211 NLP, see natural language processing No Free Lunch theorem, 10 node, 207 average degree, 208 degree, 207 indegree, 207 outdegree, 207 normalization, 51, 56 between-slide, 61 loess, 61 of mass spectra, 82 print-tip loess, 61, 76 quantile, 57, 61, 63, 75 variance stabilization, 64 within-slide, 61 NPV, see negative predictive value nuisance parameter, 229 Index NUSE, see standard error, normalized unsealed Occam's razor, 10 one-versus-all, 18 one-way analysis of variance, 18 Onto-Express, 134 overfitting, 23-25, 150 Pajek, 219 partial least squares, 16, 159 path, 209 average length, 209 PGA, see principal component analysis FOR, see polymerase chain reaction pealc detection, 82 matching, 82 quantification, 82 peptide/protein chips, perceptual mapping, see multidimensional scaling perfect match, 53 phage display, phase application, 22 learning, 22 test, 22 training, 22 validation, 22 PLIER, 64 PLM, see multi-array probe-level model PLS, see prtial least squaresl6 PM, see perfect match polymerase chain reaction, polysemy, 256 population, 42 positive predictive value, 24 PPV, see positive predictive value pre-processing, 13 precision, 258 predictor, prevalence, 24, 97 principal component analysis, 12, 15, 108, 159 probe, probe set, 53 spike-in probe set, 58 probes, 53 279 PROcess, 98, 114 profile, axray, gene expression, protein expression, projection pursuit, 113 ProteinChip, 89, 93, 114 proteomics, PubMed, 252 qRT-PCR, 3, see quantitative real-time reverse transcriptase PCR QT-Clust, 136 quantitative real-time reverse transcriptase PCR, randomization, 46 recall, 258 receiver operating characteristic, 169 reference design, 45 reference RNA, 45 regression, 10 least angle, 121 regularized logistic, 166 regularization, 155 relative log expression, 59, 76 replicate biological, 43 technical, 43 replication, 42 reverse transcriptase, ribonuclease, ribonuclease protection assay, RLE, see relative log expression RLR, see regression, regularized logistic RMA, see robust multi-chip analysis robust multi-chip analysis, 56-57, 62, ROC, see receiver operating charax;teristic RPA, see ribonuclease protection assay RProteomics, 98 S-|-ArrayAnalyzer, 62 S2N, see signal-to-noise SAGE, see serial analysis of gene expression SAM, see significance analysis of microarrays 280 Index SAM scoring criterion, 152, 169 SAMBA, 131 Sammon mapping, 111 sample, 1, 42 sampling, 173-185, 196 bootstrapping, see bootstrap k-fold random subsampling, 28 random subsampling, 27 repeated random subsampling, 178 single hold-out method, 27 split-sample, 175 two-fold nested resampling, 181 Savitzky-Golay, 90 scale-freeness, 203 scaling metric, 107 nonmetric, 109 ScanAlyze, 55 segmentation, 55 SELDI-TOF, see surface-enhanced laser desorption/ionization time-of-fiight self-organizing maps, 113, 126, 130 self-organizing tree algorithm, 126, 128, 130 sensitivity, 24, 97 serial analysis of gene expression, set learning, 25, 176 test, 25, 176 training, 191 validation, 25, 181, 191 shrunken centroid classifier, 181 signal-to-noise, 18, 59, 82, 90, 190 significance analysis of microarrays, 154 silhouette coefiicient, 131 simulated annealing, 150, 155 singular value decomposition, 15 SiZer plot, 88 small-n-large-p problem, small-world pattern, 203, 209 SOMs, see self-organizing map SOTA, see self-organizing tree algorithm specificity, 24 spectrum, Spot, 55 SSH, see suppression subtraxitive hybridization standard error normalized unsealed, 58, 75 stress function, 106 squaxed, 109 weighted, 108 study experimental, 40 observational, 40 subtractive hybridization, SUDWT, 96-97 summarization, 53 support vector machine, 11, 156, 187-200 suppression subtractive hybridization, surface-enhanced laser desorption/ionization time-of-fiight, 6, 79, 104, 161, 194 SVD, see singular value decomposition SVM, see support vector machine SVMLight, 197 SW pattern, see small-world pattern synonymy, 256 t-statistic, 17, 169 tag, taxget, test Anderson-Darling, 11 ANOVA, 18 Bartlett, 17 Benjamin! and Hochberg, 21, 233 Brown and Forsythe, 19 Cochran, 19 Duncan, 19 Dunnett, 19 F-test, 18 Hochberg, 21 Holm, 20, 134 Kruskal-WalUs, 19 Levene, 17 McNemar, 30 post-hoc, 19 random permutation, 21, 152, 153, 183 Storey and Tibshirani, 21 Student, 19 t-test, 17, 152, 169 Tukey, 19 variance-corrected resampled, 30-31 Index Welch, 19 Wilcoxon rank-sum, 152 testing, 25, 182 text mining, 32, 251-270 full text mining, 257 TIC, see total ion current time resolution, 79 time series analysis, 227-247 time-of-flight, 5, 79 TOF, see time-of-flight topological overlap analysis, 211 total ion current, 89 training, 25, 182 transcriptomics, truly alternative, 20 truly null, 20 two-dimensional difference in-gel electrophoresis, two-dimensional polyacrylamide gel electrophoresis, 281 UCSF Spot, 55 UDWT, see undecimated discrete wavelet transform undecimated discrete wavelet transform, 83 validating, 25 validation, 182 variance, 22 VSN, see normalization, variance stabilization Welch-Satterthwaite, 17 Wolfe dual, 199 wrapper, 150, 155, 160, 190 yeast two-hybrid, z-score transformation, 15 Printed in the United States .. .FUNDAMENTALS OF DATA MINING IN GENOMICS AND PROTEOMICS FUNDAMENTALS OF DATA MINING IN GENOMICS AND PROTEOMICS Edited by Werner Dubitzky University of Ulster, Coleraine, Northern Ireland Martin... References 12 Text Mining in Genomics and Proteomics Robert Hoffmann 12. 1 Introduction 12. 1.1 Text Mining 12. 1.2 Interactive Literature Exploration 12. 2 Basic Concepts 12. 2.1 Information Retrieval 12. 2.2... This is in stark contrast with conventional data mining applications in finance, retail, manufacturing and engineering, for which Daniel Berrar, Martin Granzow, and Werner Dubitzky data mining was